Stockfish Testing Queue

Finished - 20718 tests

16-10-17 Gu DoubtfulCastling diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8474 W: 1490 L: 1579 D: 5405
sprt @ 10+0.1 th 1 Penalty for castling under a potential attack
16-10-17 sn knight_forward_mobility diff
LLR: -2.65 (-2.94,2.94) [0.00,5.00]
Total: 9312 W: 1605 L: 1678 D: 6029
sprt @ 10+0.1 th 1 Take 5: bonus=S(0,12), and only for knight. Fixed bench, priority -1.
16-10-17 pb ext_night_mobility diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 3513 W: 606 L: 717 D: 2190
sprt @ 10+0.1 th 1 Take 2
16-10-17 sn knight_forward_mobility diff
LLR: -2.16 (-2.94,2.94) [0.00,5.00]
Total: 6443 W: 1124 L: 1188 D: 4131
sprt @ 10+0.1 th 1 Take 4: bonus=S(0,30), and only for knight
16-10-17 El tune_cr diff
18112/20000 iterations
36833/40000 games played
40000 @ 20+0.2 th 1 Tune connected rooks
15-10-17 sn mtg3' diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 176471 W: 31256 L: 31397 D: 113818
sprt @ 20+0 th 1 Use ratio=min(0.5, ratio), moveOverhead=30 and tweak for sudden death. Tested at time control 20+0 (sudden death, 20 seconds), without adjudication rules.
16-10-17 sg pawn_break diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 17986 W: 3215 L: 3264 D: 11507
sprt @ 10+0.1 th 1 Bonus for possible pawn break
16-10-17 sn knight_forward_mobility diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 3960 W: 657 L: 765 D: 2538
sprt @ 10+0.1 th 1 Take 3
15-10-17 sn mtg3 diff
LLR: -2.97 (-2.94,2.94) [-3.00,1.00]
Total: 44569 W: 7975 L: 8209 D: 28385
sprt @ 16+0 th 1 Use ratio=min(0.5, ratio), moveOverhead=30. Tested at time control 16+0 (sudden_death), without adjudication rules.
15-10-17 Ro QRT_Reversed diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 23314 W: 4079 L: 4106 D: 15129
sprt @ 10+0.1 th 1 Experiment: Contrary to other pieces, give more bonus for Queen threat in opponent camp instead of threats in own camp
15-10-17 xo movestogo3cfix3 diff
ELO: 11.64 +-4.5 (95%) LOS: 100.0%
Total: 10000 W: 2400 L: 2065 D: 5535
10000 @ 10/2 th 1 try fix3: instead of adding some overhead for 5 or 10 movestogo, use whole value of movestogo/2. Test and master have overhead = 20 and adjudication disabled.
15-10-17 sn mtg3 diff
ELO: -0.13 +-2.1 (95%) LOS: 45.1%
Total: 40000 W: 7301 L: 7316 D: 25383
40000 @ 10+0.1 th 1 Estimate the Elo value of using ratio=min(0.5, ratio), moveOverhead=30 at standard time control 10+0.1 . Tested without adjudication rules.
15-10-17 Ro QueenRankThreat diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 25490 W: 4467 L: 4485 D: 16538
sprt @ 10+0.1 th 1 Take 2: Smaller QRT
15-10-17 xo movestogo3cfix2 diff
ELO: 17.04 +-4.5 (95%) LOS: 100.0%
Total: 10000 W: 2479 L: 1989 D: 5532
10000 @ 10/2 th 1 Run best test against base that also includes the lower move overhead and no adjudication, so that test and base are comparable.
15-10-17 sn knight_forward_mobility diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6863 W: 1187 L: 1283 D: 4393
sprt @ 10+0.1 th 1 Take 2: generalize to bishops and knights
15-10-17 Ro QueenRankThreat diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12425 W: 2121 L: 2194 D: 8110
sprt @ 10+0.1 th 1 Compute rank threats by queen
15-10-17 II sudden_death diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 65912 W: 11739 L: 11694 D: 42479
sprt @ 10+0.1 th 1 (standard STC now) Last try: from the previous tests it is obvious that aggressive time usage is unfortunately important for Elo performance, and as this has also slight impact on increment case (should be tested separately), this is my final recommendation for possible reduction of time losses in sudden death case.
15-10-17 Ro RankThreats2 diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 19770 W: 3482 L: 3524 D: 12764
sprt @ 10+0.1 th 1 Compute rank threats only once.
15-10-17 xo movestogo3cfix2 diff
ELO: -1.36 +-4.4 (95%) LOS: 27.3%
Total: 10000 W: 2076 L: 2115 D: 5809
10000 @ 10/2 th 1 Run fix2 against itself to see how many time fails there are. (This has moveOverhead = 20 and adjudication disabled to provoke time fails, so we expect to get some.)
15-10-17 SC materialDraw diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 7711 W: 1389 L: 1482 D: 4840
sprt @ 10+0.1 th 1 Take 2 (will change the prio when take 1 finishes).
15-10-17 SC materialDraw diff
LLR: -2.94 (-2.94,2.94) [0.00,5.00]
Total: 39772 W: 7167 L: 7123 D: 25482
sprt @ 10+0.1 th 1 Do we gain Elo by checking appropriately for an insufficient material draw.
15-10-17 sn knight_forward_mobility diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 35260 W: 6251 L: 6228 D: 22781
sprt @ 10+0.1 th 1 Endgame bonus/malus for restricted knights
15-10-17 Ro DblThreats diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17391 W: 3099 L: 3150 D: 11142
sprt @ 10+0.1 th 1 Compute threats by knight and bishop separately
15-10-17 Ro DblThreats diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 8631 W: 1528 L: 1617 D: 5486
sprt @ 10+0.1 th 1 Take2: Score threat by rank only once
15-10-17 sn forward_mobility diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 2224 W: 398 L: 517 D: 1309
sprt @ 10+0.1 th 1 Forward mobility for all piece types
15-10-17 SC materialDraw diff
ELO: 0.65 +-3.2 (95%) LOS: 65.2%
Total: 15547 W: 2754 L: 2725 D: 10068
20000 @ 10+0.1 th 1 Quick check for the effect of correctly checking draw due to insufficient material.
15-10-17 An captureLMR4B diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 13414 W: 2316 L: 2385 D: 8713
sprt @ 10+0.1 th 1 Same as captureLMR4A, but do not do anything different from master for cut nodes
15-10-17 An captureLMR4A diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 8634 W: 1490 L: 1579 D: 5565
sprt @ 10+0.1 th 1 Try this again, remove the new static eval logic.
15-10-17 sn mtg3 diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 15671 W: 2759 L: 2630 D: 10282
sprt @ 60/15 th 1 Use ratio=min(0.5, ratio), moveOverhead=30. Tested at time control 60/15 (60 moves in 15secs), without adjudication rules.
15-10-17 II sudden_death diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 8636 W: 1646 L: 1504 D: 5486
sprt @ 16+0 th 1 Last try: from the previous tests it is obvious that aggressive time usage is unfortunately important for Elo performance, and as this has also slight impact on increment case (should be tested separately), this is my final recommendation for possible reduction of time losses in sudden death case.
15-10-17 sn mtg3 diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 8877 W: 1713 L: 1571 D: 5593
sprt @ 10+0.1 th 1 Use ratio=min(0.5, ratio), moveOverhead=30. Tested at time control 10+0.1, without adjudication rules.
15-10-17 El nmp_tt2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 16373 W: 2869 L: 2925 D: 10579
sprt @ 10+0.1 th 1 Last try was the best one so far. Try to improve on it.
14-10-17 xo movestogo3cfix2 diff
ELO: 8.86 +-4.5 (95%) LOS: 100.0%
Total: 10000 W: 2311 L: 2056 D: 5633
10000 @ 10/2 th 1 fix2 for the fails deliberately created in 3c. First fix looks reasonable (<1% time fails instead of 3-4%), try to improve on it. Uses movetogo values up to 10 and a double so that halves can be included. overhead 20 and adjudication disabled as before to provoke the fails.
14-10-17 El nmp_tt2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 60901 W: 10730 L: 10601 D: 39570
sprt @ 10+0.1 th 1 Use ONE_PLY offset in the other direction
14-10-17 Fi ttMoveUpdate diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 17792 W: 3136 L: 3186 D: 11470
sprt @ 10+0.1 th 1 Change how we update moves for existing TT entries. 2MB hash
14-10-17 II mtg' diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 113963 W: 20008 L: 20042 D: 73913
sprt @ 60/15 th 1 Try to reduce possibility of time losses, take 2 - cleaner solution.
14-10-17 xo movestogo3c diff
ELO: -3.54 +-4.6 (95%) LOS: 6.4%
Total: 10000 W: 2204 L: 2306 D: 5490
10000 @ 10/2 th 1 take 3c: move overhead 20 and adjudication off for more time fails.
14-10-17 pb avgdepth diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 16890 W: 3029 L: 3082 D: 10779
sprt @ 10+0.1 th 1 take 2
14-10-17 II sudden_death diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 8672 W: 1496 L: 1668 D: 5508
sprt @ 16+0 th 1 Take 2: only 5% of time after move 60 (lowered priority of the first test)
14-10-17 xo movestogo3cfix diff
ELO: 8.62 +-4.6 (95%) LOS: 100.0%
Total: 10000 W: 2411 L: 2163 D: 5426
10000 @ 10/2 th 1 take3cfix: last test (3c) had 2.3% time fails. Add mtg calc to try to reduce these.
14-10-17 pb avgdepth3 diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 5554 W: 933 L: 1035 D: 3586
sprt @ 10+0.1 th 1 take 3
14-10-17 fa pawn diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 78935 W: 14055 L: 13946 D: 50934
sprt @ 10+0.1 th 1 Tiny tweak for Pawns
14-10-17 xo movestogo3b diff
ELO: 2.99 +-4.6 (95%) LOS: 89.9%
Total: 10000 W: 2316 L: 2230 D: 5454
10000 @ 10/2 th 1 Take3b: move overhead at 30 and adjudication off to try to provoke more time fails. 10/2 similar time to 40/10 but with more time controls.
14-10-17 II mtg' diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 39446 W: 8097 L: 8334 D: 23015
sprt @ 1/0.25 th 1 Let's also see how my solution works in another extreme case.
14-10-17 El nmp_tt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 19228 W: 3390 L: 3434 D: 12404
sprt @ 10+0.1 th 1 Take 4: same as take 3, but with a ONE_PLY offset
14-10-17 sn mtg2 diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 18250 W: 3250 L: 3125 D: 11875
sprt @ 60/15 th 1 Use MoveOverhead=0, but also ratio=min(0.25,ratio). Tested at time control 60/15 and without adjudication rules.
14-10-17 sn mtg2 diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 9022 W: 1539 L: 1711 D: 5772
sprt @ 10+0.1 th 1 Use MoveOverhead=0, but also ratio=min(0.25,ratio). This way SF never uses more than 1/4th of the remaining time for one move. Tested with disabling of adjudication rules.
14-10-17 xo movestogo3 diff
LLR: -1.83 (-2.94,2.94) [0.00,5.00]
Total: 11961 W: 2219 L: 2245 D: 7497
sprt @ 40/10 th 1 movestogo3: provoke more time fails, try to fix them later
14-10-17 sn mtg2 diff
ELO: -105.94 +-8.0 (95%) LOS: 0.0%
Total: 4043 W: 525 L: 1721 D: 1797
20000 @ 10+0.1 th 1 Use MoveOverhead=0: this should bring heaps of time losses for the mtg2 branch compared to master'. Tested with disabling of adjudication rules.
14-10-17 El nmp_tt diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 28032 W: 4947 L: 4954 D: 18131
sprt @ 10+0.1 th 1 Take 3